fix(ci): clean Ascend runner workspace before checkout - #8
Open
wolala3434 wants to merge 33 commits into
Open
Conversation
…#1240) Add standard Apache 2.0 copyright header (Copyright 2026 FlagOS Contributors) to all source files. - **703 files** stamped with Apache 2.0 headers - **80 files** with upstream copyright (HuggingFace, Physical Intelligence, NVIDIA) — left untouched - **104 files** with other licenses (51 MPL, 41 proprietary, 10 MIT, 2 BSD) — third-party/upstream code - **31 files** skipped by rule — binary, JSON, etc. This PR was written in part with the assistance of generative AI.
Change tar_dir from /data/image_tar (local disk) back to /mnt/airs-business/cicd/image_tar (shared LeoFS mount), ensuring tar files are accessible from all runners.
- Change container_volumes defaults from /home/flagscale_cicd/... to /mnt/airs-business/cicd/... (shared storage) in both workflow input and run_cuda_tests job - Add set -euo pipefail and file existence check to tar save/load steps - Change file check from [ -f ] to sudo test -s for reliability - Add prepare dependency to load_images job for tar_dir reference
* ci: enable Ascend test workflow * Revert "ci: enable Ascend test workflow" This reverts commit 1573538. * ci: add Ascend image and training pipeline * ci: use manual Ascend dev image for training tests * ci: use manual Ascend dev image for training tests * fix: make Ascend install scripts executable * test: catch Typer exit exceptions * test: use Typer exit in config validation tests * fix(ci): disable CUDA-only fusions in Ascend smoke * feat(ascend): install TransformerEngine-FL for training * test(ascend): exercise full TE-FL training stack * fix(ascend): keep proxy credentials out of image metadata * fix(ascend): disable NPU autoload while building train deps * fix(ascend): align image tag with base Python * fix(ascend): use task-specific runtime base images * fix(ascend): align torchvision with torch-npu runtime * test: normalize marked inference golden output * fix(ascend): validate Megatron import before skipping install * fix(install): resume interrupted sccache downloads * fix(ascend): provide python config for dataset helpers * fix(ascend): validate TransformerEngine integration * fix(ascend): persist build download cache * test(ascend): use mock data for training smoke * fix(ascend): validate full training stack before skipping * fix(ascend): install full training stack in test env * refactor(ci): keep CLI workflow platform neutral * fix(ascend): make mock data the only smoke source * fix(ascend): omit data path from mock smoke * fix(ascend): rebuild and validate training stack * fix(ascend): initialize NPU before stack validation * fix(ascend): validate stack after platform selection * fix(ascend): make mock training smoke executable * Revert "fix(ascend): make mock training smoke executable" This reverts commit 0c50c33. * fix(ascend): restore executable mock training smoke * fix(ascend): initialize TE before Megatron platform * fix(ascend): explicitly initialize TE NPU backend * fix(ascend): reapply TE patch after Megatron setup * ci(ascend): use validated TE train image * ci(ascend): build and validate isolated task runtimes * ci(ascend): activate prebuilt training runtime * fix(ascend): isolate CANN Python paths per runtime * fix(ci): satisfy format and terminology checks --------- Co-authored-by: wolala3434 <wolala3434@users.noreply.github.com> Co-authored-by: 马函廷 <mahanting@mahantingdeMacBook-Air.local>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Add the same pre-checkout workspace cleanup used by the CUDA and MetaX image workflows to the trusted Ascend build job.
Container jobs can leave root-owned files such as
.geminiin the self-hosted runner workspace. Without this cleanup,actions/checkoutfails withEACCESbefore the image build starts.Validation
EACCES: permission denied, rmdir .../.geminitest_v0.1dispatch: https://github.com/BrianPei/FlagScale/actions/runs/30321929409